Playing with Chrome’s Canary new built-in AI. Made a detective story (with plot holes) where the Gemma 4 plays detective and transcribes evidence recording, analyses images etc - all offline.
Lives here https://t.co/0DzNa67glC
Made myself a Wallpaper Studio to keep my wallpapers constantly updated. Most of the time I use local ComfyUI with different models, but also I use headless Grok or Codex.
So far Codex is bringing the best gens, but the goal is to achieve close enough results with local models.
In 2023, they said we had 5 years. In 2024, they said we should learn biology instead. In 2025, they said AI would write 100% of the code within a year. Today... well, it's still "6 to 12 months" away.
Here is https://t.co/kFLD4AcjoQ so we know when to pivot.
@sultanbegova Привет из НЗ. Тут тоже своих проблем хватает. По идее все те же самые проблемы, что и везде, масштаб другой немного.
Жизнь дорогущая, преступность растет, правительство не фонтан - короче все как везде, но есть и плюсы - кругом красиво, тварей ядовитых нет.
@NousResearch's Hermes Agents WebUI Dashboard is finally taking shape.
It's taking a bit longer than expected because I need to surface both Hermes agents (from the PC and the Raspberry Pi) in one unified view - slowly, but surely.
Question into void as probably nobody will see it, but what's the best model so far to run locally on 4070 12Gb.
Purpose: hermes agent - so it would need to be very capable of tool calling.
I'm trying Gemma 4 right now, but it's slow as on my 12Gb...
@pcshipp I'm running Codex cli in VS Code in the ull editor view, but was thinking to try Wave. Periodically checking diffs is the only real thing sopping me from fully switching to CLI.